Hi Jeremiah Willcock, Attached is the zip package of boost.mpi, pyhton, parallel_graph building and testing logs. Environment is MinGW (gcc 4.7.1) boost.mpi and parallel_graph could be built, but none of the graph_parallel could be built. Thanks Zhiyu Li 2013-07-01 lizy10b 发件人： Jeremiah Willcock 发送时间： 2013-06-29 07:35:53 收件人： boost-users 抄送： 主题： Re: [Boost-users] Confused with Depth-First-Searchmethod,forwardorcross edge found on a undirected graph ? On Sat, 29 Jun 2013, lizy10b wrote:
Hi Jeremiah Willcock , I have tested r84912 and r84913:
r84912 -- built failed (line 319) Error 3 error C2668: 'boost::detail::bfs_helper' : ambiguous call to overloaded function e:\boost\binary_include\include\boost-1_53\boost\graph\breadth_first_search.hpp 319 1 boost_test_1
r84913 -- built success So that is taken care of, then? I knew you would need both patches; sorry for the extra work of checking with only r84912.
And for the building boost.mpi with mingw, you sugguest me "try adding MPICH_SKIP_MPICXX as a -D flag (to define it as a macro) when building Boost.MPI". sorry, I dont fully understand, does it mean I should try to add some flag (MPICH_SKIP_MPICXX as a -D or something) when execute the b2 command to build boost.mpi? You need to somehow add -DMPICH_SKIP_MPICXX to your compiler flags; I don't remember how to do that with Boost.Build, but you should be able to look it up. -- Jeremiah Willcock
___________________________________________________________________________________________________________________________________________________________ lizy10b
___________________________________________________________________________________________________________________________________________________________ 发件人： Jeremiah Willcock 发送时间： 2013-06-29 03:56:57 收件人： boost-users 抄送： 主题： Re: [Boost-users] Confused with Depth-First-Searchmethod,forwardorcross edge found on a undirected graph ? On Fri, 28 Jun 2013, lizy10b wrote:
Hi Jeremiah Willcock
I download the two breadth_first_search.hpp files from the trunk (now is R84910) and overwrite my existing files repectiviely. But unfortunately VS2010 complains about one error when try to build.
Error 3 error C2668: 'boost::detail::bfs_helper' : ambiguous call to overloaded function e:\boost\binary_include\include\boost-1_53\boost\graph\breadth_first_search.hpp 319 1 boost_test_1
Double click the error message will let the mouse point to Line 319 Could you please try the changes in r84912 and r84913? They affect the files: boost/graph/distributed/concepts.hpp boost/graph/graph_traits.hpp boost/graph/breadth_first_search.hpp boost/graph/distributed/breadth_first_search.hpp -- Jeremiah Willcock
________________________________________________________________________________________________________________________________________________________
lizy10b
________________________________________________________________________________________________________________________________________________________
Hi Jeremiah Willcock, Good news, the parallal version BFS example (\graph_parallel\example\breadth_first_search.cpp) could be built now. I followed your suggestion to locate the boost::detail::bfs_helper(), there are two declarartions of this function, A and B. A is in \graph\breadth_first_search.hpp for the serial version, with another definition for the parallel version of this function (B) which is wrappe d inside a macro (#ifdef BOOST_GRAPH_USE_MPI... #endif). And the delcaration of B is in \graph\distributed\breadth_first_search.hpp. I just comment out the delcaration of B and copy it to overwrite the definition inside the above macro. and append "* = 0" at the end of the last parameter just like function A. And then the example cpp built successfully. Could you please see if r84909 in the trunk (just committed) works without needing any extra changes? I added the "= 0" there but didn't do any of
发件人： Jeremiah Willcock 发送时间： 2013-06-28 00:54:20 收件人： boost-users 抄送： 主题： Re: [Boost-users] Confused with Depth-First-Searchmethod,forwardorcross edge found on a undirected graph ? On Fri, 28 Jun 2013, lizy10b wrote: the other changes you mention.
As for the build error of Boost.MPI with MinGW, I will redo it and save the ouput to a file. Thanks. -- Jeremiah Willcock
______________________________________________________________________________________________________________________________________________________ __ _ lizy10b
______________________________________________________________________________________________________________________________________________________ __ _ 发件人： Jeremiah Willcock 发送时间： 2013-06-27 22:47:02 收件人： boost-users 抄送： 主题： Re: [Boost-users] Confused with Depth-First-Searchmethod,forwardorcross edge found on a undirected graph ? On Thu, 27 Jun 2013, lizy10b wrote:
Hi Jeremiah Willcock, Thanks for your reply. Sad to say the pararallel version BFS example file (\boost_1_53_0_\libs\graph_parallel\example\breadth_first_search.cpp) can not be compiled with VS 20 10 on my end. It complains an unresolved external symbol "void _cdecl boost::bfs_helper....". And I found a similar report on http://boost.2283326.n4.nabble.com/LINK-Error-while-building-PBGL-example-br.... but with no result. Is it boost::bfs_helper or boost::detail::bfs_helper? Are there any symbols in the object file for breadth_first_search that are defined and demangle to names involving bfs_helper? I also tried to build the boost with MinGW, but it seems the Boost.MPI cannot be compiled correctly upon MPICH2, so the parallel BGL library will no t be built either. What is the issue with that version? -- Jeremiah Willcock
2013-06-27
____________________________________________________________________________________________________________________________________________________ __
lizy10b
____________________________________________________________________________________________________________________________________________________ __
Hi Jeremiah Willcock, Thanks for your reply. As maybe there are some problems when invoke depth_first_visit() on an undirected graph, the undirected_dfs() could be a very good substitution. So I continue to test the parallel version of the DFS function -- tsin_depth_first_visit(), and the test file comes from \boost_1_53_0\libs\graph_parallel\test\distributed_dfs_test.cpp. The test file invokes the tsin_depth_first_visit() on a distributed undirected graph, and the visitor print all events, the "forward_or_cross_edge" appears four times. So maybe there are some problems in tsin_depth_first_visit() too. Does the parallel tsin_depth_first_visit() base on the serial depth_first_visit(), so they appear similiar problem when work with undirected graph? It looks like it's similar to the directed DFS, so it would likely return
发件人： Jeremiah Willcock 发送时间： 2013-06-26 22:54:58 收件人： lizy10b 抄送： boost-users 主题： Re: Re: Re: [Boost-users] Confused with Depth-First-Searchmethod,forwardorcross edge found on a undirected graph ? On Wed, 26 Jun 2013, lizy10b wrote: the same results.
My final target is to find a parallel version DFS works with undirected graph, in order to find all cycles parallelly. Could you please give me some advices on that? I am stucked. It doesn't look like there is one, but you can use breadth-first search to find cycles as well. Would that work for you? -- Jeremiah Willcock
____________________________________________________________________________________________________________________ lizy10b
____________________________________________________________________________________________________________________ 发件人： Jeremiah Willcock 发送时间： 2013-06-26 03:34:37 收件人： lizy10b 抄送： boost-users 主题： Re: Re: Re: [Boost-users] Confused with Depth-First-Searchmethod,forwardor cross edge found on a undirected graph ? On Tue, 25 Jun 2013, lizy10b wrote:
Hi Jeremiah Willcock, Can I use the predecessor_map in undirected_dfs() like in dijkstra_shortest_paths()? undirected_dfs(g, predecessor_map(&p[0]).root_vertex(vertex_descriptor(0)).visitor(vis) .edge_color_map(get(edge_color, g)).vertex_color_map(get(vertex_color, g))); It does not work on my end, all elements in predecessor remain 0 after invoke the undirected_dfs() . If that parameter isn't in the documentation for undirected_dfs, the argument will just be ignored. In the last section (Extending Algorithms with Visitors) of http://www.boost.org/doc/libs/1_53_0/libs/graph/doc/q uick_tour.html, an example shows how to write a predecessor_map using visitor, and the "event point" is edge_relaxed(). But as the undirected_dfs document shows there is no edge_relaxed() for DFS. So is there any way to save predecessors in DFS? For example an edge (u,v) and start at vertex u When the discover_vertex(Vertex, Graph&) is invoked at vertex v, then inside the discover_vertex() how could I know the previous discovered vertex is u? I believe you would want to hook tree_edge() instead of edge_relaxed(). -- Jeremiah Willcock
_________________________________________________________________________________________________________________
发件人： Jeremiah Willcock 发送时间： 2013-06-25 00:44:05 收件人： lizy10b 抄送： boost-users 主题： Re: Re: [Boost-users] Confused with Depth-First-Search method,forwardor cross edge found on a undirected g raph ? On Tue, 25 Jun 2013, lizy10b wrote: > Hi Jeremiah Willcock , > ? > It seems undirected_dfs() works. I have not燾heck the result according to the graph, but at least there is no > forward edge or cross edge output. > Attached is the revised cpp file using undirected_dfs() . > My question is why edge_color_map(get(edge_color,爂))?is required when calling the undirected_dfs() . > Thanks It looks like what is going wrong with using directed DFS is that edges are examined twice (once in each direction), and so you can have forward edges even in a symmetric graph (which is how undirected graphs are treated in directed-graph algorithms in BGL). An example is the triangle 0 -- 1 -- 2 -- 0: starting at 0 and processing the edges out of a vertex in numerical order, 0 -> 1 is a tree edge, 1 -> 0 is a back edge, 1 -> 2 is a tree edge, 2 -> 0 is a back edge, 2 -> 1 is a back edge, then 0 -> 2 is a forward edge. An undirected DFS would mark 2 -> 0 as a back edge then ignore 0 -> 2 since it is the same as 2 -> 0. The edge colors in undirected_dfs are used to do this filtering-out of multiple directions for the same edge. -- Jeremiah Willcock
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users